## Monday, November 21, 2011

### How to Lie with Statistics

Perusing my favorite blogs, I ran across this graph from Stephen Pinker's latest book:

The general charge seems to be that "war deaths" are more prolific in stateless societies than the state-based ones we're familiar with. Is anyone else ready to pull their hair out over the misleading nature of that graphic? Let's think about this for a second.

I'll go ahead and assume that the "stateless" examples indeed represent truly stateless societies, and that they are fairly representative of war within those societies (i.e. not cherry-picked examples). I want to make this argument as strong as possible before I unravel it. Notice that the metric being used is "deaths per 100,000 people." Alright, so now we're not talking about some nominal value - we're talking about a ratio or percentage. So, for this, we need the nominal number of deaths for some sample (n). I won't even bother to contest the number of deaths. I'm not a historian. I'm assuming Pinker did his homework. We'll assume that what's provided is correct. But what about the sample (n) being used?

It's stuff like this that drives me up a wall. Notice that the "stateless societies" in question are much smaller and that their wars are more regional and isolated. One might expect, when talking about a regional conflict in which the total amount of people, land, and resources more generally is much smaller, then the cost of war (as a percentage of those total resources) is going to be much higher, relatively speaking.

Contrast this with nationalistic endeavors, where the war may be fought regionally and yet largely un-involved or un-utilized resources will drive your sample (n) value up. The result, of course, is that the inflated sample makes your total cost seem low relative to smaller conflicts. Look at things this way - if two neighboring towns went to war, all of the resources, in terms of people, would be right in the middle of the fight so to speak. It would be hard to escape such conflict. So if two towns with a few thousand people each go to arms over something, having 1,000 people die in the conflict would be devastating according the the statistical analysis above.

On the other hand, in the case of, say, WWII, literally hundreds of thousands of American soldiers can perish in the battlefields of Europe, but since there are 150 million other Americans back in the homeland who aren't even remotely near such regional conflict, the total percentage of Americans killed in war seems small - despite the obvious and incredible nominal disparity. Likewise the astonishing number of deaths occurring in large nation-states like Russia or Japan all of a sudden seem almost insignificant when bundled with their massive and largely uninvolved populace (and I mean uninvolved as in actually warring for contested territory, etc.). Notice that the ratio seems more significant for each of the "states" listed as the proximity between each of them and the regions of involved conflict during the sample period tighten.

To illustrate the issue I see a little further, let's think about how this analysis/comparison would apply in another scenario:

Let's say that there was a city in Utah called "Secessionville" that tried to break ties with both the state and federal governments. And, although this is not how it would happen, let's assume that the state and federal governments, independently, sought recourse through militaristic means. Secessionville barricades itself and prepares for war. The state readies whatever regiment it decides to put together for the purpose of reclaiming the town and the federal government, separately, does likewise in their own interest(s).

So here we have three separate groups about to engage in a regional war. Secessionville has 50,000 people. Utah has about 3,000,000 people. The United States has roughly 310,000,000 people. After three long weeks of battle, federal and state troops are (somewhat) victorious. 35,000 of those hailing from Secessionville have perished before giving up. 200,000 of those hailing from Utah (proper) have somehow managed to meet their maker. And 6.5 million very inept soldiers for the U.S. government have fumbled their way into their graves. Using statistical analysis similar to what we've seen above, how would this look?

I don't know about you guys, but it seems pretty irrefutable to me that the death toll rings much larger for smaller governments when it comes to conflict. So to all you proponents of smaller government, from minarchy to anarchy, put away your Friedman and Rothbard. You've put up a good fight but my charts, metrics, and hidden assumptions have asploded your precious deductive reasoning. What say ye now, deniers of science?!

Yeah. Exactly.